A counterexample to modus ponenses Matthew Mandelkern∗ March 10, 2020 Penultimate draft; to appear in The Journal of Philosophy Abstract McGee (1985) argued that modus ponens was invalid for the natural language conditional ‘If::: then::: ’. Many subsequent responses have argued that, while McGee’s examples show that modus ponens fails to preserve truth, they do not show that modus ponens fails to preserve rational full acceptance, and thus modus ponens may still be valid in the latter informational sense. I show that when we turn our attention from indicative conditionals (the focus of most of the literature to date) to subjunctive conditionals, we find that modus ponens does not preserve either truth or rational full acceptance, and thus is not valid in either sense. In concluding I briefly consider how we can account for these facts. 1 Introduction McGee (1985) argued that modus ponens was invalid for the natural language conditional ‘If::: then::: ’ on the basis of examples like the following: Opinion polls taken just before the 1980 election show the Republican Ronald Reagan decisively ahead of the Democrat Jimmy Carter, with the other Republican in the race, John Anderson, a distant third. Those apprised of the poll results believed, with good reason: (1) If a Republican wins the election, then if it’s not Reagan who wins it will be Anderson. (2) A Republican will win the election. Yet they did not have reason to believe (3) If it’s not Reagan who wins, it will be Anderson. ∗All Souls College, Oxford, OX1 4AL, United Kingdom; [email protected]. Thanks to an anony- mous referee for this journal, Justin Bledin, David Boylan, Kevin Dorst, Kurt Norlin, Richard Roth, Paolo Santorio, and Ginger Schultheis for very helpful comments on this paper. 1 A series of responses since have pointed out that, though examples like this certainly show something interesting about modus ponens, it is not clear exactly what it is. In particular, these responses have argued that, while cases like this might show that modus ponens is invalid in the sense that it does not preserve truth, these cases do not challenge modus ponens as a principle about preservation of full acceptance (sometimes called informational consequence). Indeed, it looks like, when you fully accept (1) and (2), then you are required to fully accept (3), and so examples like this are no counterexample to the informational version of modus ponens. In this paper I will argue that, while this response seems correct for indicative conditionals, the focus of most of the literature, it does not extend to subjunctive conditionals: in the case of subjunctives, both kinds of modus ponens, truth- conditional and informational, are invalid. 2 Two kinds of modus ponens I begin by explaining the two kinds of modus ponens that have been distinguished in response to McGee’s examples. My take on the basic idea is this.1 McGee’s counterexamples show that you can have high credence in pp, and if p, then qq without being logically compelled to have high credence in q (since you can have high credence in the conjunction of (1) and (2) without being compelled to have high credence in (3)). This suggests that modus ponens is not truth-preserving: if q is true in any case where pp and if p, then qq are both true, then, assuming rational credence can be modeled probabilistically, the laws of probability would rule out this state of affairs.2 So McGee’s counterexample is a good counterexample to the rule that says that the inference from p and pIf p, then qq to q is truth-preserving. Call this rule truth-conditional modus ponens. Truth-conditional modus ponens is one natural way of spelling out what modus ponens amounts to, and McGee’s examples indeed appear to show that it is not valid. Nonetheless, there is a different way of spelling out modus ponens which these examples do not refute: namely, as the principle that if you fully accept both p and pIf p, then qq, then you are rationally 1The basic idea, in differing formulations, goes back at least to Over 1987. I am here following most closely discussion in Santorio 2020. 2I will assume that conjunction has classical properties which allow us to move freely between the two-premise formulation of modus ponens and the corresponding formulation with one conjunctive premise. This simplifies discussion but is not essential. 2 required to fully accept q. Call this rule informational modus ponens. On reflection, McGee’s case does not look like a counterexample to informational modus ponens. After all, in McGee’s case, if you really fully accept that a Republican will win, in the sense of fully ruling out that a Democrat won (suppose a perfectly reliable oracle tells you that a Republican will win, or a perfectly reliable newspaper prints ‘Carter loses!’), then you must indeed accept that, if it’s not Reagan who wins, it’s Anderson. For if you get conclusive evidence that a Republican will win, without learning which Republican it will be, then you leave open just two options: Reagan and Anderson. And so it seems that you ought to accept that, if it’s not Reagan, it’s Anderson. This line of response is spelled out in detail in a number of places. Bledin (2015) in particu- lar (building on work in Veltman 1996; Gillies 2004, 2009; Yalcin 2007) proposes a conception of logic on which it is ‘a science that is fundamentally concerned not with the preservation of truth but rather with the preservation of structural properties of the bodies of information that we generate, encounter, absorb, and exchange as we interact with one another and learn about our world’. On that informational conception, ‘an argument is logically valid if and only if any body of information that incorporates all of its premises also incorporates its conclusion by virtue of logical form’. When we think in these terms, McGee’s examples indeed do not threaten the validity of modus ponens, for the reason just given: if you fully incorporate the premises into your information, then you ought to incorporate the conclusion as well.3 In sum, then, it seems that there are (at least) two reasonable ways of saying precisely what modus ponens amounts to: 1. Truth-conditional modus ponens: the inference from p together with pIf p, then qq to q always preserves truth. 2. Informational modus ponens: the inference from p together with pIf p, then qq to q always preserves rational full acceptance. McGee’s counterexamples show the first principle is invalid, but they do not show that the second principle is invalid. 3There are different ways that the notion of informational consequence could be spelled out, but differences in detail won’t matter for our purposes. Cf. Norlin 2019 for related discussion; see also Stern and Hartmann 2018 for a slightly different, but related, distinction between two kinds of modus ponens. 3 3 The problem with subjunctives The literature on modus ponens has focused on indicative conditionals (like the conditional in (1)), rather than subjunctives (like ‘If a Republican had won, then if it hadn’t been Reagan, it would have been Anderson’). There is a natural reason for this: as McGee points out, it is difficult to assess modus ponens for conditionals in the subjunctive mood, because subjunctives tend to communicate that their antecedents are false (e.g. ‘If the picnic had been cancelled, Susie would have been upset’ generally communicates that the picnic wasn’t cancelled).4 Levi (1996, p. 106), Brogaard and Salerno (2008), and Briggs (2012) have attempted to formulate counterexamples to modus ponens in the subjunctive mood, but the results are not compelling, to my ear, for precisely the reason McGee points out. Here is Briggs’s case, for illustration: A firing squad (consisting of two executioners, X and Y) is about to shoot a prisoner. If the court orders the execution, this will cause the captain to signal to the executioners. Each executioner is posed to fire if, and only if, the captain signals. Either executioner’s shot by itself will be enough to kill the prisoner. As a matter of fact, the court does order the execution, so that the captain signals, the executioners shoot, and the prisoner dies. Briggs then considers this conditional: (4) If executioner X had fired, then (even) if the captain had not signalled, the prisoner would have died. The thought then seems to be that (4) and ‘Executioner X fired’ can both be true, while ‘(even) if the captain had not signalled, the prisoner would have died’ is not. It feels hard to assess this claim, however, because it is hard to assess (4) in a context like the one given, where we know that Executioner X fired. In that case—when we know that the antecedent is true—my intuitions about (4) feel very unclear. So, while I am sympathetic to Briggs’s aims, I do not think Briggs’s case yields very clear judgments. Similar points go for Levi’s and Brogaard and Salerno’s cases, which have the same structural problem.5 4McGee speculates that modus ponens is likely invalid for subjunctives, since import-export appears to be valid for subjunc- tives, and there is a tension between modus ponens and import-export drawn out by Gibbard (1981); however, see Etlin 2008 for potential counterexamples to import-export for subjunctives which make this line of argument less convincing. 5A complication about evaluating modus ponens for subjunctives concerns mood. In English, there will often be morpho- 4 4 Subjunctive modus ponens Importantly, however, not all subjunctives suggest that their antecedents are false.